Hey everyone! 
Wanted to share a project I’ve been hacking on: a SpaceMouse-driven teleoperation and data collection pipeline for the AgileX PiPER 6-DOF arm on ROS 2.
The main idea was to make teleoperation feel natural instead of clunky. You just grab a 6-axis SpaceMouse, steer the gripper through Cartesian space, and collect clean demonstration datasets (packaged for LeRobot / ACT imitation learning).
A few quick details on how it’s set up:
- Controller: SpaceMouse 6-DOF inputs mapped to Cartesian velocities with joint-space smoothing filters so the arm doesn’t jerk around.
- The Sim Twin: I hooked up a live MuJoCo digital twin alongside the hardware. Right now, honestly, it’s mostly there for visuals and nice rendering (you could just as easily use RViz for the basic joint transforms!).
- What’s next: The real reason to keep MuJoCo around is the next step—adding synthetic camera views in the sim that mirror my physical camera setup. That way, I can test and benchmark trained vision policies in simulation before letting the model loose on the physical robot table.
The repo is public, and I’d love any thoughts, feedback, or ideas on this project.